kubegc/uni-share

GPU Manager

question

1、若构建失败将脚本以及一些非代码文件转换格式

apt install dos2unix
dos2unix <filename>

deploy

1、

#在所有有GPU的节点上打上标签
kubectl label node <你的GPU节点> nvidia-device-enable=enable
#在master节点上复制yamlfile文件夹中的文件到指定文件夹
cp scheduler-extender.yaml /etc/kubernetes
cp scheduler-policy-config.json /etc/kubernetes

2、修改kube-scheduler.yaml文件

• 备份kube-scheduler.yaml文件

cp /etc/kubernetes/manifests/kube-scheduler.yaml ./

• command处增加

- command:
  - --config=/etc/kubernetes/scheduler-extender.yaml

• volumeMounts处增加

volumeMounts:
- mountPath: /etc/localtime
      name: localtime
      readOnly: true
- mountPath: /etc/kubernetes/scheduler-extender.yaml
      name: extender
      readOnly: true
- mountPath: /etc/kubernetes/scheduler-policy-config.json
      name: extender-policy
      readOnly: true

• volumes处增加

volumes:
- hostPath:
      path: /etc/localtime
      type: File
    name: localtime
- hostPath:
      path: /etc/kubernetes/scheduler-extender.yaml
      type: FileOrCreate
    name: extender
- hostPath:
      path: /etc/kubernetes/scheduler-policy-config.json
      type: FileOrCreate
    name: extender-policy

3、

//cd deploy
//chmod +x gpuDeploy.sh
//./gpuDeploy.sh